This paper presents a generative adversarial network (GAN) model for the restoration of Dunhuang murals, addressing the challenges of incomplete restoration and local detail distortion. The proposed model combines a parallel dual convolutional feature extraction depth generator and a ternary heterogeneous joint discriminator. The generator, designed with vanilla convolution and dilated convolution, captures multi-scale features to reduce information loss. The pixel-level discriminator, along with a global and local discriminator, ensures detailed and accurate restoration. The method is validated on a Dunhuang murals dataset, showing improved performance in PSNR and SSIM metrics compared to existing methods. The restored images are more visually appealing and align better with human perception, demonstrating effective restoration of mural images. The key contributions include the parallel dual convolutional feature extraction module and the pixel-level discriminator, which enhance the overall quality and detail of the restored images.This paper presents a generative adversarial network (GAN) model for the restoration of Dunhuang murals, addressing the challenges of incomplete restoration and local detail distortion. The proposed model combines a parallel dual convolutional feature extraction depth generator and a ternary heterogeneous joint discriminator. The generator, designed with vanilla convolution and dilated convolution, captures multi-scale features to reduce information loss. The pixel-level discriminator, along with a global and local discriminator, ensures detailed and accurate restoration. The method is validated on a Dunhuang murals dataset, showing improved performance in PSNR and SSIM metrics compared to existing methods. The restored images are more visually appealing and align better with human perception, demonstrating effective restoration of mural images. The key contributions include the parallel dual convolutional feature extraction module and the pixel-level discriminator, which enhance the overall quality and detail of the restored images.